JPH10111792A - Picture processor - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide picture compression/reproduction devices which can execute a program optimum for the processing performance of present CPU and DSP in accordance with multiple CPU and DSP. SOLUTION: The picture compression device reads pictures and compresses an I picture being an intra-frame code, a P picture being an inter-fram code only in a forward direction and a B picture being an inter-frame code in forward/backward directions in a picture compression means 44. At that time, a device control means 41 checks present CPU 42, refers to a module table 56 and selects a module optimum for the processing performance of CPU 42 from the program 55. The optimum modules are combined and the optimum program is generated. The picture reproduction device executes the processing similar to the picture compression device and generates the optimum program.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an apparatus for compressing and encoding an image code by a CPU, a DSP, or the like, and an apparatus for reproducing a compressed image code.

[0002]

2. Description of the Related Art When an image is digitized and recorded on a recording medium such as a CD-ROM or a hard disk, the data amount is enormous, so that it is usually recorded by compression encoding. Among various compression coding methods, a DCT-based coding method that performs compression by utilizing the property that the spatial frequency of an image is concentrated at a low frequency is relatively frequently used.
This is JPEG (Joint Photographi)
c Expert Group) or MPEG (Movi)
ng Pictures Expert Group)
And other international standard coding methods.

FIG. 1 shows a hierarchical diagram of a code format conforming to MPEG. MPEG codes have several hierarchical structures as shown in FIG. The top layer is a video sequence, and includes a plurality of GOPs (Group O).
f Picture).

[0004] A GOP is composed of a plurality of pictures.
One picture indicates one image. There are three types of pictures: an I picture which is an intra-frame code, a P picture which is an inter-frame code only in the forward direction, and a B picture which is a bi-directional inter-frame code. A picture is composed of a plurality of slices divided into arbitrary regions. A slice is composed of a plurality of macroblocks arranged in left-to-right or top-to-bottom order.

A macroblock is a luminance component (Y1, Y2, Y3, Y4) obtained by further dividing a 16 × 16 dot block into 8 × 8 dot blocks, and a color difference of an 8 × 8 dot block in an area corresponding to the luminance component. It is composed of six blocks of components (Cb, Cr). An 8 × 8 dot block is the minimum unit of encoding.

The compression of an image code by a conventional DCT-based coding method will be described with reference to the drawings using MPEG as an example. FIG. 2 is a block diagram of an image compression apparatus that compresses an image code conforming to MPEG. The image compression apparatus shown in FIG.
The motion data is converted into UV data, and the motion search means 6 searches for the motion of the image of the previous / next frame and the current frame for each 8 × 8 block area. Next, in MPEG, the intra-frame code I
Since there are three types of compression: a picture, a P picture, which is an inter-frame code only in the forward direction, and a B picture, which is a bi-directional inter-frame code, the three types of compression are performed.

[0007] In the case of an I picture, the DCT means 8 performs discrete cosine transform on the pixel values of the 8 × 8 block area of the current frame, quantizes the result by the quantizing means 9, and the variable length coding means 10. High-efficiency compression to Huffman code Next, in order to decode the compressed image as a reference frame, the quantized data is inversely quantized by the inverse quantization means 14, and inverse discrete cosine transform is performed by the IDCT means 13 to calculate a pixel value. Stored in the unit 11.

In the case of a P picture, the motion prediction means 7
Thus, the pixel value of the 8 × 8 block area of the current frame and the pixel value of the 8 × 8 block area of the previous frame stored in the reference frame unit 11 referred to by the motion searched by the motion search means 6 The difference is calculated, the difference value is subjected to discrete cosine transform by the DCT means 8, quantized by the quantizing means 9, and the variable-length coding means 10 performs high-efficiency compression to a variable-length Huffman code. Next, in order to decode the compressed image as a reference frame, the quantized data is inversely quantized by the inverse quantization means 14 and the IDCT means 1
3 to calculate the difference value by the inverse discrete cosine transform, and the motion compensating unit 12 calculates the pixel value of the 8 × 8 block area of the previous frame stored in the reference frame unit 11 referred to by the motion predicting unit 7. The difference values are added and stored in the reference frame unit 11.

In the case of a B picture, the motion prediction means 7
Accordingly, the pixel values of the 8 × 8 block area of the current frame and the pixels of the 8 × 8 block area of the previous / next frame stored in the reference frame unit 11 referred to by the motion searched by the motion search means 6 are calculated. The difference is calculated by the DCT means 8 and the difference value is subjected to the discrete cosine transform.
, And the variable-length encoding means 10 efficiently compresses the variable-length Huffman code. Since the B picture is not used as a reference frame, the image is not expanded.

The reproduction of an image code according to a conventional DCT-based coding method will be described with reference to the drawings using MPEG as an example. FIG. 3 is a block diagram of an image reproducing apparatus for reproducing an image code conforming to MPEG.
The image reproducing apparatus shown in FIG. 3 reads three codes and reads three kinds of codes: an I picture which is an intra-frame code, a P picture which is an inter-frame code only in the forward direction, and a B picture which is a bi-directional inter-frame code. To stretch.

In the case of an I picture, the variable length decoding means 25
, And inversely quantized by the inverse quantization means 26,
At T27, the values of the pixels of the block are calculated by the inverse discrete cosine transform, and the image is output by the RGB conversion means 28.

In the case of a P picture, the picture is decoded by the variable length decoding means 25 and inversely quantized by the inverse quantization means 26.
The IDCT 27 calculates the difference between the blocks by the inverse discrete cosine transform, adds the difference to the motion-compensated block of the previous frame stored in the reference frame unit 29 by the motion compensation unit 30, and outputs an image by the RGB conversion unit 28. I do.

Further, in the case of a B picture, it is decoded by the variable length decoding means 25 and inversely quantized by the inverse quantization means 26.
The IDCT unit 27 calculates the difference between the blocks by the inverse discrete cosine transform, and the motion compensating unit 30 performs the motion-compensated block of the previous frame stored in the reference frame unit 29 and the previous / next frame stored in the reference frame unit 29. Is added to the motion-compensated block, and an image is output by the RGB conversion means 28.

If the compression and reproduction are performed based on the international standard MPEG, the image can be compressed and reproduced with high efficiency. However, motion search / motion compensation and DCT
Because many operations are required for processing such as / IDCT, image compression / reproduction by software needs to be performed by a program optimized to operate at the highest speed in consideration of the characteristics of the CPU. Many types of C
In order to support PU, it is necessary to prepare and switch different programs.

As a conventional example using an optimum program, Japanese Patent Laid-Open Publication No. Hei 6-282444 discloses a method of compiling a program so as to be optimal for a CPU. Japanese Patent Application Laid-Open No. 4-322329 discloses a method of creating an optimum program at the time of installation on a computer.
An example of selecting an optimal CPU according to a program is disclosed in Japanese Patent Application Laid-Open No. 3-240467.

[0016]

However, in the above-described conventional optimization program creating method, it is necessary to create a dedicated program for each computer to be used and install it on the computer to be used. And can not share the program. In addition, since the program is not created in consideration of the amount of available memory, the program may be too large to be loaded into the memory. In addition, the optimal CPU for the program
Has a drawback that it cannot be applied to a computer operating with a single CPU.

An object of the present invention is to provide an image compression / playback apparatus which operates a program most suitable for the current CPU and memory at the time of executing the program.

[0018]

According to the present invention, there is provided an image processing apparatus for encoding an input image, comprising:
A module table describing modules to be loaded for each module, a program for managing a plurality of modules specified by the module table, device control means for controlling loading of the modules into the memory, and loaded modules. An image processing apparatus characterized by having an image compression unit is obtained.

According to the present invention, there is provided an image processing apparatus for expanding an input code, comprising: a module table describing a module to be loaded for each CPU; and a plurality of modules specified by the module table. An image processing apparatus is provided which has a management program, device control means for controlling loading of a module into a memory, and image reproducing means including the loaded module.

[0020]

Embodiments of the present invention will be described with reference to the drawings.

FIG. 4 is a block diagram of the image compression apparatus of the present invention. In FIG. 4, a device control means 41 for controlling the whole apparatus, a CPU 42 for operating a program, a keyboard 43 for transmitting an input from a user, and an image compression means 4 for performing image compression
4 and a program 55 storing an optimum module for each CPU of each module, and a module table 56 showing each CPU and the optimum module corresponding thereto. The image compression unit 44 includes a YUV conversion unit that converts an image into YUV data, a motion search unit 46 that searches for the motion of the image, a motion prediction unit 47 that calculates a difference between blocks from the motion of the image, and a discrete cosine transform. DCT means 48 for transforming, quantizing means 49 for quantizing the value after discrete cosine transform, variable length coding means 50 for encoding to Huffman code, and inverse quantizing means 54 for inverse quantizing the value after quantization. IDCT means 53 for performing an inverse discrete cosine transform of the value after inverse quantization, motion compensation means 52 for adding a difference value to a block of a motion-compensated reference frame, and a reference frame unit 51 for storing preceding and succeeding reference frames. I have.

The image compression apparatus shown in FIG.
The UV conversion means 45 converts the data into YUV data, and the motion search means 46 searches for the motion of the image of the previous / next frame and the current frame for each 8 × 8 block area. Next, in MPEG, there are three types: an I picture which is an intra-frame code, a P picture which is an inter-frame code only in the forward direction, and a B picture which is a bi-directional inter-frame code. Compression is performed.

In the case of an I picture, the DCT unit 48 performs a discrete cosine transform of the pixel values of the 8 × 8 block area of the current frame, quantizes the quantized unit 49, and the variable length encoding unit 50 performs variable length coding. High-efficiency compression to Huffman code Next, in order to decode the compressed image as a reference frame, the quantized data is
, And inverse discrete cosine transform by the IDCT means 53 to calculate a pixel value.
To be stored.

In the case of a P picture, the motion prediction means 4
7, the value of the pixel of the area of the 8 × 8 block of the current frame and the value of the pixel of the area of the 8 × 8 block of the previous frame stored in the reference frame unit 51 referred to by the motion searched by the motion search means 46 , The DCT means 48 performs a discrete cosine transform of the difference value, quantizes the quantization value with the quantization means 49, and highly efficiently compresses the data into a variable length Huffman code by the variable length coding means 50. Next, in order to decode the compressed image as a reference frame, the quantized data is inversely quantized by the inverse quantization
The difference value is calculated by the inverse discrete cosine transform by the T means 53, and 8 of the previous frame stored in the reference frame unit 51 referred to by the motion prediction means 47 by the motion compensation means 52.
The difference value is added to the value of the pixel in the area of the × 8 block and stored in the reference frame unit 51.

In the case of a B picture, the motion prediction means 4
7, the pixel value of the 8 × 8 block area of the current frame and the 8 × 8 of the previous / next frame stored in the reference frame unit 51 referred to by the motion searched by the motion search means 46.
The difference is calculated based on the pixel values in the area of 8 blocks, and the DCT is calculated.
The difference value is subjected to discrete cosine transform by the means 48, quantized by the quantizing means 49, and compressed by the variable length coding means 50 into a variable length Huffman code with high efficiency. In the case of a B picture, since it is not used as a reference frame, the process of creating the reference frame unit 51 after the inverse quantization 54 is not performed.

The device control means 41 is provided with the current CPU 4
2, the module necessary for configuring the optimal program for the CPU 42 is obtained from the module table 56, and the optimal module is combined to create an optimal program.

FIG. 5 is a block diagram of an image reproducing apparatus according to the present invention. In FIG. 5, device control means 6 for controlling the entire device
1, a CPU 62 for operating a program, a keyboard 63 for transmitting an input from a user, an image reproducing means 64 for reproducing an image, a program 71 storing an optimal module for each CPU of each module, and an optimal module corresponding to each CPU Is constituted by a module table 72 indicating the following. The image reproducing means 64 includes a variable length decoding means 65 for decoding the compression code, an inverse quantization means 66 for inversely quantizing the decoded value, and an IDCT means 67 for inverse discrete cosine transform of the inversely quantized value. It comprises an RGB conversion unit 68 for converting YUV data into RGB data, a reference frame unit 69 for storing reference frames before and after, and a motion compensation unit 70 for calculating a difference between blocks from the motion of an image.

The image reproducing apparatus shown in FIG. 5 reads a code and reads an I-picture which is an intra-frame code, a P-picture which is an inter-frame code only in the forward direction, and a B-picture which is a bi-directional inter-frame code. Decompress the type sign.

In the case of an I picture, the variable length decoding means 65
, And inversely quantized by the inverse quantization means 66,
At T67, the pixel values of the block are calculated by the inverse discrete cosine transform, and the image is output by the RGB conversion means 68.

Further, in the case of a P picture, the picture is decoded by the variable length decoding means 65 and inversely quantized by the inverse quantization means 66.
The IDCT 67 calculates a block difference by inverse discrete cosine transform, adds the difference to the motion-compensated block of the previous frame stored in the reference frame unit 69 by the motion compensation unit 70, and outputs an image by the RGB conversion unit 68. I do.

In the case of a B picture, the picture is decoded by the variable length decoding means 65 and inversely quantized by the inverse quantization means 66.
The IDCT means 67 calculates the difference between the blocks by the inverse discrete cosine transform, and the motion compensating means 70 performs the motion-compensated block of the previous frame stored in the reference frame unit 69 and the previous / next frame stored in the reference frame unit 69. Is added to the motion-compensated block, and an image is output by the RGB conversion means 68.

The device control means 61 is provided with the current CPU 6
2, the module necessary for configuring the optimal program for the CPU 62 is obtained from the module table 72, and the optimal module is combined to create the optimal program.

Next, the creation of an optimal program will be described with reference to image reproduction.

FIG. 6 is a diagram showing the flow of data when an optimum program is created. In FIG. 6, the processing capacity of the CPU is Z
In this case, the image reproduction program 81 is optimized.
First, the optimum variable length decoding module for the CPUZ is determined from the module table 83. Since the optimal variable length decoding module for the CPUZ is the variable length decoding C, the variable length decoding C
Load 1 Thereafter, similarly, inverse quantization D, IDC
The TG, motion compensation K, and RGB conversion I modules are loaded to create an image reproduction program optimal for the CPUZ.

The program 82 includes a CPU in the program.
It is composed of modules that require a large load. In the image reproduction, it is a module for variable length decoding, inverse quantization, IDCT, motion prediction, and RGB conversion.

In the case of image compression, similarly, YUV conversion,
Motion search, motion prediction, DCT, quantization, variable length coding,
Modules for inverse quantization, IDCT, and motion compensation are stored.

Next, the operation of the embodiment of the present invention will be described with reference to the drawings.

FIG. 7 is a flow chart of the device control means of the block diagram of FIG. In FIG. 7, C of the current CPU 42 is
The PUID and the amount of available memory are checked (step 91), and an optimal module for performing image reproduction with the current processing capacity is determined. Here, when it is determined that the memory amount is insufficient (step 92), the processing for the memory shortage operates (step 96). If the memory is sufficient, the optimal module obtained as a result of the determination is loaded and a compression program is created (step 93). Next, image compression is performed (step 94), and it is determined whether or not all frames have been completed (step 95). If not, the process returns to step 94.

FIGS. 8 and 9 are flowcharts of image compression. In these figures, the image is converted into YUV data by the YUV conversion means 45 (step 101), and the motion of each macroblock of the image is searched by the motion search means 46 (step 102) to determine the type of picture. (Step 103), processing is performed for each picture type.

In the case of an I picture, the DCT means 48 performs a discrete cosine transform (step 104), and the quantization means 49
(Step 105), the variable-length encoding means 50 highly efficiently compresses the variable-length Huffman code (step 106), and the quantized data is inversely quantized by the inverse quantization means 54 (step 106). 107), IDC
The inverse discrete cosine transform is performed by the T means 53 and stored in the reference frame unit 51 (step 108).

In the case of a P picture, the motion prediction means 47 calculates a difference between the block of the current frame and the block of the previous frame of the reference frame section 11 (step 109).
DCT means 48 performs discrete cosine transform (step 11).
0), quantized by the quantization means 49 (step 11)
1) Highly efficient compression into a variable length Huffman code by the variable length encoding means 50 (step 112), and the quantized data is inversely quantized by the inverse quantization means 54 (step 113), and the IDCT means 53 (Step 114), and the motion compensation unit 52 adds the difference value to the block of the previous frame of the reference frame unit 51 and stores the result in the reference frame unit 51 (step 11).
5).

In the case of a B picture, the motion prediction means 47 calculates the difference between the block of the current frame and the block of the previous / next frame of the reference frame unit 11 (step 11).
6) The discrete cosine transform is performed by the DCT unit 48 (step 117), the quantization is performed by the quantization unit 49 (step 118), and the variable length coding unit 50 compresses the data into a variable length Huffman code with high efficiency (step 117). 119), the quantized data is inversely quantized by the inverse quantization means 54 (step 120), and inverse discrete cosine transform is performed by the IDCT means 53 (step 121), and the motion compensation means 52
Thus, the difference value is added to the blocks of the previous / next frame of the reference frame unit 51 and stored in the reference frame unit 51 (step 122).

FIG. 10 is a flowchart for creating a compression program. In FIG. 9, the module is loaded using the current CPUID and YUV conversion as arguments (step 13).
1) The module is loaded with the current CPUID and the motion search as arguments (step 132), and the current CPUI
The module is loaded with D and motion prediction as arguments (step 133), the module is loaded with the current CPUID and DCT as arguments (step 134), and the module is loaded with the current CPUID and quantization as arguments ( (Step 135) The module is loaded with the current CPUID and the variable length encoding as arguments (Step 13).
6) The module is loaded using the current CPUID and the inverse quantization as arguments (step 137), and the current CPUI
The module is loaded with D and IDCT as arguments (step 138), and the module is loaded with the current CPUID and motion compensation as arguments (step 139).

FIGS. 11 and 12 are flow charts of processing when the memory is insufficient. In these figures, the current CPU
The module is loaded with the ID and the YUV conversion as arguments (step 141), and the YUV conversion is executed (step 142). After the execution, the memory of the data and the program used in the YUV conversion is released (step 143), and the current C
The module is loaded using the PUID and the motion search as arguments (step 144), and the motion search is executed (step 145). After the execution, the memory of the data and the program used in the motion search is released (step 146), and the current CP is released.
The module is loaded using the UID and the motion prediction as arguments (step 147), and the motion prediction is executed (step 1).
48). After the execution, the memory of the data and the program used in the motion prediction is released (step 149), and the current CPU
The module is loaded with the ID and the DCT as arguments (step 150), and the DCT is executed (step 15).
1). After the execution, the memory of the data and the program used in the DCT is released (step 152), and the current CPU ID
Then, the module is loaded with the arguments of the quantization and the quantization (step 153), and the quantization is executed (step 154). After the execution, the memory of the data and the program used in the quantization is released (step 155), the module is loaded with the current CPUID and the variable length encoding as arguments (step 156), and the variable length encoding is executed (step 156). Step 15
7). After the execution, the memory of the data and the program used in the variable length coding is released (step 158), and the current CP is released.
The module is loaded using the UID and the inverse quantization as arguments (step 159), and the inverse quantization is executed (step 1).
60). After the execution, the memory of the data and the program used in the inverse quantization is released (step 161), and the current CPU
The module is loaded using the ID and the IDCT as arguments (step 162), and the IDCT is executed (step 1).
63). After the execution, the memory of the data and the program used in the IDCT is released (step 164), and the current CPU
The module is loaded with the ID and the motion compensation as arguments (step 165), and the motion compensation is executed (step 1).
66). After the execution, the memory of the data and program used in the motion compensation is released (step 167), and it is determined whether or not all the frames have been completed (step 168). If not, the process returns to step 141.

FIG. 13 is a flow chart of the apparatus control means in the block diagram of FIG. In FIG. 13, the current CPU 62
The CPUID and the available memory amount are checked (step 171), and an optimal module for performing image reproduction with the current processing capacity is determined. Here, when it is determined that the memory amount is insufficient (step 172),
The memory shortage process operates (step 176). If the memory is sufficient, the optimal module obtained as a result of the determination is loaded and a reproduction program is created (step 17).
3). Next, image reproduction is performed (step 174), and it is determined whether or not all frames have been completed (step 17).
5) If not, return to step 174.

FIG. 14 is a flowchart for creating a reproduction program. In FIG. 14, the module is loaded using the current CPUID and variable length decoding as arguments (step 181), and the module is loaded using the current CPUID and inverse quantization as arguments (step 182).
The module is loaded with the PUID and IDCT as arguments (step 183), and the module is loaded with the current CPUID and RGB conversion as arguments (step 18).
4) The module is loaded using the current CPUID and motion compensation as arguments (step 185).

FIG. 15 is a flowchart of a process when the memory is insufficient. In FIG. 15, the module is loaded using the current CPUID and variable length decoding as arguments (step 191), and the variable length decoding is executed (step 19).
2). After the execution, the memory of the data and the program used in the variable length decoding is released (step 193), and the current CP is released.
The module is loaded with the UID and the inverse quantization as arguments (step 194), and the inverse quantization is executed (step 1).
95). After the execution, the memory of the data and program used in the inverse quantization is released (step 196), and the current CPU
The module is loaded using the ID and the IDCT as arguments (step 197), and the IDCT is executed (step 1).
98). After the execution, the memory of the data and the program used in the IDCT is released (step 199), and the current CPU
The module is loaded using the ID and the RGB conversion as arguments (step 200), and the RGB conversion is executed (step 201). After execution, the memory of the data and program used in the RGB conversion is released (step 202), and the current C
The module is loaded using the PUID and the motion compensation as arguments (step 203), and motion compensation is performed (step 204). After the execution, the memory of the data and the program used in the motion compensation is released (step 205), and it is determined whether or not all the frames have been completed (step 206). If not, the process returns to step 191.

FIG. 16 is a flowchart of image reproduction. In FIG. 16, the compressed code is decoded by the variable length decoding means 65 (step 211), the type of picture is determined (step 212), and processing is performed for each picture type.

In the case of an I picture, inverse quantization is performed by the inverse quantization means 66 (step 213), inverse discrete cosine transformation is performed by the IDCT means 67, and the result is stored in the reference frame section 69 (step 214). By YU
V is converted to RGB (step 215).

In the case of a P picture, inverse quantization is performed by the inverse quantization means 66 (step 216), inverse discrete cosine transform is performed by the IDCT means 67 (step 217), and the previous frame of the reference frame section 69 is moved by the motion compensation means 70. The difference value is added to the reference block and stored in the reference frame unit 69 (step 218).
V is converted to RGB (step 219).

In the case of a B picture, inverse quantization is performed by the inverse quantization means 66 (step 220), and inverse discrete cosine transform is performed by the IDCT means 67 (step 221). The difference value is added to the block of the subsequent frame, stored in the reference frame unit 69 (step 222), and converted from YUV to RGB by the RGB conversion unit 68 (step 223).

As in the case of compressing an image of a TV conference,
When a certain processing speed is required for the compression program, a plurality of CPUs having different processing speeds are loaded with an optimal module for the CPU to be used so that the required speed standard is satisfied. You need to create a program. FIG. 17 is a diagram illustrating an example of a module table of an optimal module to be loaded. In FIG.
A DCT module, a quantization module, and a variable-length encoding module are shown as an example among modules constituting the compression program. In particular, the case where the modules to be loaded are divided by focusing only on the processing speed of the CPU has been described.

First, the CPU is roughly divided into a 16-bit CPU and a 32-bit CPU, and the processing speed of each is within the range of the reference value set by the compression program or lower than the reference value. It is classified into three types, "standard", "low speed", and "high speed", depending on whether it is high speed or high speed.

A quantization module for a 16-bit CPU will be described. A quantization module for performing a truncation process is loaded when the processing speed of the CPU is low, and a quantization module for performing high-quality processing when the processing speed is high. Is loaded. If the processing speed is within the range of the reference value, a reference module that does not perform special processing is loaded.

Similarly, DCT and variable length coding modules are selected and loaded according to the processing speed of the CPU. When the CPU is 32-bit, the module created for 32-bit is selected and loaded, as in the case of 16-bit. For example, when the CPU is 32 bits and the processing speed is high, the DCT is loaded with a “high quality” module, the quantization is loaded with a “high quality” module, and the variable length coding is loaded with a “standard” module. Is done.

FIG. 18 shows a part of the quantization module for 16 bits and 32 bits. When the CPU is 32-bit, operating a 16-bit program generally lowers the operating speed as compared with a 32-bit program, and also compares FIG. 18 (A) with FIG. 18 (B). As will be understood, the number of instructions for performing the same operation is larger in the 16-bit program. Since a 16-bit CPU does not operate a 32-bit program, a program other than a 16-bit program cannot be selected. As described above, in order to switch the module to be loaded for each type of CPU, the CPU ID which is an individual value for each CPU is checked, and a module most suitable for the currently used CPU is loaded.

The termination of the processing performed by the quantization module when the processing speed of the CPU is low will be described. FIG. 19 shows an 8 × 8 block. The block after the DCT processing has been converted into a frequency component.
Are arranged from the low-frequency component to the high-frequency component in the order shown in FIG. In the standard quantization process, it is necessary to perform multiplication and division on all 64 components. However, the higher the frequency, the higher the probability that the value will be 0 after quantization. Also, even if the high frequency component is omitted, it is difficult for the human eye to see the difference. When quantizing high frequency components, the processing time can be shortened by stopping the operation at an intermediate point without performing all the 64 points of the operation. If you stop the calculation,
All parts after the stop point are filled with zeros. The cutoff point can be arbitrarily changed according to the processing speed of the CPU. If the quantization is discontinued at 32 points, the processing is completed in half the time required for performing the 64-point quantization.

Next, a description will be given of the high image quality processing of the quantization module when the processing speed of the CPU is high. To improve the image quality, it is necessary to improve the calculation accuracy. An example of this is shown in FIG. In the example, (18 ÷ 5) + (8 ÷
A program for performing the calculation of 3) is shown. If it is calculated to the first decimal place, it becomes 6.3. In the example of FIG.
The division of the 0022 line gives 18 ÷ 5 = 3, the division of the 0026 line gives 8 ÷ 3 = 2, and when these are added, the result is 5. On the other hand, FIG.
In order to reduce the effect of truncation during division,
The number divided by the 0022th line and the 0027th line is multiplied by 16, and the result after all operations are completed is divided by 16. The division on the 0023th line gives 18 × 16 ÷ 5 = 57, and the division on the 0028th line gives 8 × 16 ÷ 3 = 42.
To divide by 16 at line 0030, (57 + 42) ÷ 16 =
It becomes 6. When comparing the results, it is better to perform high-precision arithmetic,
It can be seen that the error is small because the ratio of truncation below the decimal point is small. However, the amount of calculation is increased by the amount of the process of raising the number once divided when performing the division and returning the number after the calculation. As described above, if the image quality is improved by increasing the calculation accuracy, the processing time increases. Therefore, the module is loaded as a CPU module having a high processing speed.

In order to operate a program in a state where the available memory is small, it is necessary to reduce the number of programs and data to be loaded at one time. FIG. 21 shows an example in which variable length coding is performed after quantization. For example, a data table for quantization requires 4 bytes per point, and if two tables with a total of 64 points are used, 992 bytes are required. If an attempt is made to load the encoding table in an attempt to perform variable-length encoding while the table is loaded on the memory, but there is a possibility that the table cannot be loaded due to insufficient memory, as shown in FIG. Before loading the quantization module into the memory, it is necessary to release the area used by the quantization module to increase the available memory. By repeating this for each module load, the compression program can operate even with a small memory.

Similarly, by performing YUV conversion, motion compensation, motion prediction, DCT, variable length coding, inverse quantization, and IDCT, an optimum module is selected according to the processing capacity of the CPU and the amount of memory. Can be done.

[0061]

As described above, by creating a plurality of modules constituting the compression / reproduction program, a program optimal for the processing capacity of the CPU to be used operates.

Even a program that cannot be loaded with the entire program due to insufficient memory can be operated by sequentially loading the memory into modules and discarding it after use.

[Brief description of the drawings]

FIG. 1 is a hierarchy diagram of a code format conforming to the MPEG system.

FIG. 2 is a block diagram of a conventional image compression device.

FIG. 3 is a block diagram of a conventional image reproducing apparatus.

FIG. 4 is a block diagram of an image compression device to which the present invention has been applied.

FIG. 5 is a block diagram of an image reproducing apparatus to which the present invention is applied.

FIG. 6 is a diagram showing a data flow when creating an optimal program.

FIG. 7 is a flowchart for explaining the operation of the device control means of FIG. 4;

FIG. 8 is a flowchart illustrating an image compression process.

FIG. 9 is a flowchart illustrating an image compression process.

FIG. 10 is a flowchart illustrating a compression program creation process.

FIG. 11 is a flowchart illustrating a process performed when memory is insufficient.

FIG. 12 is a flowchart illustrating a process performed when memory is insufficient.

FIG. 13 is a flowchart for explaining the operation of the device control means of FIG. 5;

FIG. 14 is a flowchart illustrating a reproduction program creation process.

FIG. 15 is a flowchart illustrating a process performed when memory is insufficient.

FIG. 16 is a flowchart illustrating an image reproduction process.

FIG. 17 is a diagram illustrating an example of a module table.

FIG. 18 is a diagram illustrating an example of a quantization program module.

FIG. 19 is a diagram showing a scanning order of pixels in an 8 × 8 block.

FIG. 20 is a diagram illustrating an example of a truncation process in a quantization program module.

FIG. 21 is a diagram showing an example of a map of a program loaded on a memory.

FIG. 22 is a diagram showing an example of a map of a program loaded on a memory.

[Explanation of symbols]

 41 Device control means 42 CPU 43 Keyboard 44 Image compression means 45 YUV conversion means 46 Motion search means 47 Motion prediction means 48 DCT means 49 Quantization means 50 Variable length coding means 51 Reference frame unit 52 Motion compensation means 53 IDCT means 54 Reverse Quantization means 55 program 56 module table 61 device control means 62 CPU 63 keyboard 64 image reproduction means 65 variable length decoding means 66 inverse quantization means 67 IDCT means 68 RGB conversion means 69 reference frame section 70 motion compensation means 71 program 72 module Table 81 Image playback program 82 Program 83 Module table

Claims (4)

[Claims] 1. An image is divided into small blocks, orthogonal transformation is performed for each block, the result of the transformation is quantized, and a highly efficient coded intra-frame code and an image are divided into small blocks. For each block, a search is made for a block that has the smallest difference between the current frame and the frames before and after the current frame (motion search), and the difference between the block of the current frame and the block of the motion searched frame is calculated. Performs orthogonal transformation on the difference block, quantizes the transformation result,
In an image processing apparatus for compressing and encoding an image with a highly efficient encoded inter-frame code, various types of CPUs or D
An image processing apparatus comprising means for creating an optimal image compression program for a current CPU or DSP from a module of an optimal image compression program for each SP. 2. The image processing apparatus according to claim 1, wherein an image is compressed by loading the data into a memory for each processing unit, executing the data, discarding the data from the memory, and repeating the processing. 3. An image is divided into small blocks, orthogonal transformation is performed for each block, the result of the transformation is quantized, and a highly efficient encoded intra-frame code and the image are divided into small blocks. For each block, a search is made for a block that has the smallest difference between the current frame and the frames before and after the current frame (motion search), and the difference between the block of the current frame and the block of the motion searched frame is calculated. In an image processing apparatus that performs DCT on the difference block, quantizes the result of the conversion, and reproduces an image obtained by compression-encoding the image with the highly efficient inter-frame code, a variety of C
An image processing apparatus comprising: means for creating an optimal image reproduction program for a current CPU or DSP from a module of an optimal image reproduction program for each PU or DSP. 4. The image processing apparatus according to claim 3, wherein each of the processing units is loaded into a memory, executed, discarded from the memory, and an image is reproduced by repeating the processing.